Modern multi-behavior recommendation system frameworks have been greatly improved, and many researchers have made contributions to enable the recommendation models to capture more hidden information. Nevertheless, we consider that previous multi-behavior recommendation system(MBR) models still could not well address two major challenges: (1) solving the category imbalance problem between target behaviors and auxiliary behaviors; (2) considering and correctly handling the commonalities between different users. Hence, we propose an Adaptive Multi-view Multi-behavior Contrastive Learning Recommendation(AMMCR), consisting of a sequential view and a graph view. We design one adaptive fusion module, two supervised learning tasks and multi-behavior contrastive learning layer for existing challenges in both views to solve above challenges. The Adaptive fusion module and Adaptive Supervised task aim to consider varying degrees of importance between target behaviors and auxiliary behaviors. The Self-supervised task attempts to concentrates on the coarse-grained commonality between different users. The multi-behavior contrastive layer aims at optimizing representations of different behaviors and different views. In experiments, our model outperforms existing SOTA baselines. Source code of our work is available at https://github.com/Anticoder1/AMMCR .